Evolving artificial neural networks for short term load forecasting
Introduction
Short-term load forecasting is aimed at predicting electric loads for a period of minutes, hours, days or weeks. Short-term load forecasting plays an important role in the real-time control and the security functions of an energy management system. Throughout the years, considerable research efforts have been devoted to short-term load forecasting and the enhancement of its overall accuracy pertaining to the different days and events of the year. So far, principal methods that have surfaced in the field of short-term load forecasting include time-series prediction methods and regression methods 2, 4, 7. The former necessitates the analysis of historical data for an effective projection of future characteristics while the latter relies on a strong correlation of relevant variables to electric load. Time-series models employ the historical data for extrapolation to obtain future hourly loads. The disadvantage of these models is that the load trend is stationary and that weather information or any other factors that contribute to the load behaviour cannot be fully utilised. Regression models analyse the relationship among loads and other influential factors such as weather and customer usage behaviour. The main disadvantage is that these models require complex modelling techniques and heavy computational efforts to produce reasonably accurate results
The emergence of artificial intelligence techniques in recent years have seen an enormous upwelling of interest in their application to industrial processes. Their advantage lies in the fact that they do not require any complex mathematical formulations or quantitative correlation between inputs and outputs. Many years' data are also not necessary. Effective utilisation of artificial intelligence in the context of ill-defined processes have led to their application in load forecasting. Consequently, pattern recognition [6], expert systems 14, 9and neural networks 3, 10, 11, 17have been proposed for electric load forecasting. Expert-system based methods capture the expert knowledge into a comprehensive database which is then used for predicting the future load. These models are discrete and logical in nature, and use the knowledge of a human expert to develop rules for forecasting. However, transforming the knowledge of an expert to a set of mathematical rules is often very difficult.
The artificial neural network-based models have been found to be the most popular for load forecasting applications. The advantage of ANN over statistical models lies in its ability to model a multivariate problem without making complex dependency assumptions among input variables [18]. Furthermore, the ANN extracts the implicit non-linear relationship among input variables by learning from training data. The critical issue in applying neural networks and other data-driven forecasting systems is generalisation, the performance on data not used for training. The key to generalisation behaviour is model complexity. Too simple a model cannot approximate the true relationship, and overly complex models adjust to the noise in the data. Most applications of non-parametric forecasting models (such as neural networks) vary model complexity by adjusting the number of parameters. Although in recent years, artificial neural networks have been successfully applied in electric load forecasting problem, finding the optimal network structure and weights of the neural network for best results has been a challenging issue. Nearly all studies in this field apply cross-validation to select the best model. Of particular interest are the statistical properties (i.e., bias and variance) of model selection methods in estimating out-of-sample performance.
A relatively new area of research involves the hybrid fuzzy neural approach (FNN) to load forecasting. Several collaborative combinations of the fuzzy controllers with neural networks have been explored 15, 1, 12, 5, 16, including the fuzzy pre-processor for neural inputs, the fuzzy post-processor for neural outputs, the integrated fuzzy-neural network and the parallel fuzzy-neural forecaster. The use of neural networks for training eliminates the need for large historical database. Besides, the load forecaster is easily updated by frequent retraining with new data. When the situation arises whereby additional factors are found to affect the load demand, changes and inclusions can be easily incorporated into the fuzzy rules to enable the load forecaster to adapt to the new conditions. The ease with which modifications can be made to this load forecaster whenever changes occur is a feature that cannot be found in load forecasters based on traditional methods such as time series prediction and regression. Results reported from these combinations have been encouraging, with forecasting accuracy higher than those obtainable by the conventional methods.
In recent years, significant advancement has been made in the field of genetic algorithms (GAs). Genetic algorithm-based load forecasting methods 19, 8, 13have been reported to yield results that have been more than encouraging.
The majority of models that use artificial neural networks (e.g. the most frequently used back-propagation model) have a set of problems:
- •
Dependence on initial parameters.
- •
Long training time.
- •
Lack of problem-independent way to choose appropriate network topology.
- •
Incomprehensive (black box) nature of ANNs.
Section snippets
GA for function optimisation
Genetic algorithms are based on models of genetic change in a population of individuals. These models consist of three basic elements: a fitness measure which governs an individual's ability to influence future generations, a selection and reproduction process which produces offspring for the next generation and genetic operators which determine the genetic makeup of the offspring. The distinguishing feature of a GA with respect to other function optimisation techniques is that the search
The evolved ANN model
The performance of an ANN depends not only on the input variable selections but also on the network dimension. ANN will not learn and generalize with insufficient network size.
The model proposed for the load forecasting problem is a three-layered feed-forward back-propagation network trained by the genetic algorithm as described above. At the beginning of the training process, the architecture of the model was fixed as shown in Fig. 1 and the weights between the layers were allowed to adjust
Data, target variables and performance metrics
In the model, the forecast temperature for the next hour, and the average load values of the next and the current hours were used to predict the load for the next hour. Although the forecast is obtained on an hour to hour basis, the load values are obtained for the whole day using the model.
For the forecasts of weekdays i.e. Tuesday, Wednesday, Thursday and Friday, there is one network for each of these days. For the forecast of Saturdays and Sundays, the model uses two neural networks each –
Validation and results
The training and forecast for the model was performed on the actual load and temperature data of a year's data (1 January 1995–30 December 1995) obtained from a local utility. The training time varied depending on the input variables and the speed at which the genetic algorithm searched for an optimum solution. The stopping criteriona for the genetic algorithm was set such that the training would stop the moment the fitness function did not change for 30 generations. This indicated that the
Conclusion
This paper has presented an experiment which evolves single hidden layer neural networks for application in one day ahead electric load forecasting, demonstrating the usefulness of genetic algorithms in a practical application. The actual weather and load data obtained from a utility were used for training and testing of the networks. The neural networks, trained and evolved by the GA were shown to have a strong advantage over a statistical method currently used by the utility. Finally, this
Dipti Srinivasan (Member, IEEE) is an Assistant Professor with the Department of Electrical Engineering, National University of Singapore. Her research interest is in the application of soft computing techniques in power system operation, economics and control.
References (19)
- et al.
Forecasting monthly electric load and energy for a fast-growing utility using an artificial neural network
Electric Power Systems Research
(1995) - et al.
Short-term load forecasting using neural networks
Electric Power Systems Research
(1995) - et al.
Short-term load forecasting by a neural network and a refined genetic algorithm
Electric Power Systems Research
(1994) - et al.
A novel neuro-fuzzy based self-correcting online electric load forecasting model
Electric Power Systems Research
(1995) - et al.
A neural network short term load forecaster
Electric Power Systems Research
(1994) - et al.
Short term load forecasting using fuzzy neural networks
IEEE Trans. Power Syst.
(1995) - et al.
Comparative Models for Electrical Load Forecasting
(1985) - et al.
Weather sensitive short-term load forecasting using nonfully connected artificial neural network
IEEE Trans. Power Syst.
(1992) Short-term load forecasting using general exponential smoothing
IEEE Trans. Power Appar. Syst. PAS-
(1971)
Cited by (58)
SVR-FFS: A novel forward feature selection approach for high-frequency time series forecasting using support vector regression
2020, Expert Systems with ApplicationsCitation Excerpt :Prediction of electric demand in residential areas was studied in STLF by Son and Kim (2015), who combined SVR and fuzzy-rough feature selection with particle swarm optimization. Several studies used artificial neural networks (ANN) for this task with mixed results, mostly due to the known overfitting issue (Hippert, Pedreira, & Souza, 2001; Lee, Cha, & Park, 1992; Srinivasan, 1998). Traditional methods such as autoregressive moving average (Huang & Shih, 2003) and exponential smoothing (Christiaanse, 1971) are still used instead to face this challenge thanks to their simplicity and good predictive performance.
Optimal sub-models selection algorithm for combination forecasting model
2015, NeurocomputingCitation Excerpt :Forecasting has always been an essential instrument in science research, and has received much attention during the past three decades [1–4]. The main types of forecasting models include ARIMA models [5,6], exponential smoothing models [7,8], neural networks models [9,10], and support vector regression models [11–13]. Some researchers in this area often study the nature of an individual model, and try to identify the best one to generate a forecast.
Imperialist competitive algorithm combined with refined high-order weighted fuzzy time series (RHWFTS-ICA) for short term load forecasting
2013, Energy Conversion and ManagementCitation Excerpt :The load data are very intricate; therefore, accurate forecasting is not easy for STLF [5]. In recent decades, a large number of STLF methods (e.g., the Support Vector Regression (SVR) [6–8], Artificial Neural Network (ANN) [9–13], fuzzy logic [14,15]) have been proposed by the research community to achieve an accurate forecast. Recently, evolutionary algorithms (EAs) have received a large amount of attention from researchers and have been considered to be useful in many applications [16–33] and specially to the load forecasting area [4,8,34–36].
Short-term Maharashtra state electrical power load prediction with special emphasis on seasonal changes using a novel focused time lagged recurrent neural network based on time delay neural network model
2011, Expert Systems with ApplicationsCitation Excerpt :Kalaitzakis, Stavrakakis, and Anagnostakis (2002) proposed parallel implementations of multilayer perceptron neural networks (MLPs). Three layered feed forward back propagation networks (BPNs) trained by genetic algorithm is suggested by Srinivasan (1990). Ho, Hsu, and Yang (1992) give comparison of adaptive learning algorithm over conventional BP algorithm.
Application of neuro-fuzzy networks to forecast innovation performance - The example of Taiwanese manufacturing industry
2010, Expert Systems with ApplicationsApplication of data models in power grids for loss reduction and disaster anticipation
2023, E3S Web of Conferences
Dipti Srinivasan (Member, IEEE) is an Assistant Professor with the Department of Electrical Engineering, National University of Singapore. Her research interest is in the application of soft computing techniques in power system operation, economics and control.